the virtual kidney: an escience interface and grid portal · the virtual kidney uses a web...

doi: 10.1098/rsta.2008.0291, 2141-2159367 2009 Phil. Trans. R. Soc. A

Robert Moss, William Appelbe, Peter J. Hunter and S. Randall ThomasPeter J. Harris, Rajkumar Buyya, Xingchen Chu, Tom Kobialka, Ed Kazmierczak, portalThe Virtual Kidney: an eScience interface and Grid

Referencesl.html#ref-list-1http://rsta.royalsocietypublishing.org/content/367/1896/2141.ful

This article cites 33 articles, 4 of which can be accessed free

Subject collections

(22 articles)computer modelling and simulation � (51 articles)computational biology �

collectionsArticles on similar topics can be found in the following

Email alerting service herein the box at the top right-hand corner of the article or click

Receive free email alerts when new articles cite this article - sign up

http://rsta.royalsocietypublishing.org/subscriptions go to: Phil. Trans. R. Soc. ATo subscribe to

This journal is © 2009 The Royal Society

on 4 May 2009rsta.royalsocietypublishing.orgDownloaded from

http://rsta.royalsocietypublishing.org/content/367/1896/2141.full.html#ref-list-1

http://rsta.royalsocietypublishing.org/cgi/collection/computational_biology

http://rsta.royalsocietypublishing.org/cgi/collection/computer_modelling_and_simulation

http://rsta.royalsocietypublishing.org/cgi/alerts/ctalert?alertType=citedby&addAlert=cited_by&saveAlert=no&cited_by_criteria_resid=roypta;367/1896/2141&return_type=article&return_url=http://rsta.royalsocietypublishing.org/content/367/1896/2141.full.pdf?ijkey=eY1ufRTGmqZvGZK&keytype=finite

http://rsta.royalsocietypublishing.org/subscriptions

http://rsta.royalsocietypublishing.org/


The Virtual Kidney: an eScience interfaceand Grid portal

BY PETER J. HARRIS1,*, RAJKUMAR BUYYA

2, XINGCHEN CHU2,

TOM KOBIALKA2, ED KAZMIERCZAK

2, ROBERT MOSS2,

WILLIAM APPELBE3, PETER J. HUNTER

4AND S. RANDALL THOMAS

5

1Faculty Information Technology Unit, Faculty of Medicine,Dentistry and Health Sciences, and 2Department of Computer Science and

Software Engineering, The University of Melbourne, Victoria 3010, Australia3Victorian Partnership for Advanced Computing, Carlton South,

Victoria 3053, Australia4Bioengineering Institute, University of Auckland, Auckland 1142, New Zealand5IBISC (Informatiques, Biologie Integree et Systemes Complexes) CNRS FRE

3190, Universite d’Evry, Val d’Essonnes, 91000 Evry, France

The Virtual Kidney uses a web interface and distributed computing to provideexperimental scientists and analysts with access to computational simulations andknowledge databases hosted in geographically separated laboratories. Users can explorea variety of complex models without requiring the specific programming environment inwhich applications have been developed.This initiative exploits high-bandwidth communication networks for collaborative

research and for shared access to knowledge resources. The Virtual Kidney has beendeveloped within a specialist community of renal scientists but is transferable to otherareas of research requiring interaction between published literature and databases,theoretical models and simulations and the formulation of effective experimental designs.A web-based three-dimensional interface provides access to experimental data, a

parameter database and mathematical models. A multi-scale kidney reconstructionincludes blood vessels and serially sectioned nephrons. Selection of structures provideslinks to the database, returning parameter values and extracts from the literature.Models are run locally or remotely with a Grid resource broker managing scheduling,

monitoring and visualization of simulation results and application, credential andresource allocation. Simulation results are viewed graphically or as scaled colourgradients on the Virtual Kidney structures, allowing visual and quantitativeappreciation of the effects of simulated parameter changes.

Keywords: kidney modelling; computational biology;three-dimensional anatomical visualization; Grid computing; Physiome;

cyberinfrastructure

On

*A

Phil. Trans. R. Soc. A (2009) 367, 2141–2159

doi:10.1098/rsta.2008.0291

e contribution of 15 to a Theme Issue ‘The virtual physiological human: tools and applications II’.

uthor for correspondence ([email protected]).

2141 This journal is q 2009 The Royal Society


P. J. Harris et al.2142


1. Background

(a ) The Physiome and virtual physiological human

The Kidneyome project, also known as the ‘Virtual Kidney’, is part of the largerInternational Union of Physiological Sciences Physiome Project, an internation-ally collaborative open source project to provide a public domain framework forcomputational physiology, including the development of modelling standards,computational tools and web-accessible databases of models of structure andfunction at all spatial scales: interactions; protein pathways; integrative cellfunction; tissue and whole-organ structure–function relationships (Thomas et al.2008; Thomas in press); and finally the integrative function of the wholeorganism (Bassingthwaighte 2000; Hunter & Borg 2003). A key initiative forpromoting and supporting development of the Physiome in the EuropeanSeventh Framework Programme is the virtual physiological human (VPH),which has been defined as ‘a methodological and technological framework that,once established, will enable collaborative investigation of the human body as asingle complex system’. VPH is not ‘the supermodel’ that will explain all possibleaspects of human physiology or pathology. It is a way to share observations,derive predictive hypotheses from them and integrate them into a constantlyimproving understanding of human physiology/pathology, by regarding it as asingle system (Fenner et al. 2008).

The Virtual Kidney represents an information management model that can beimplemented for other organ systems within the Physiome Project or the VPH(or, indeed, for non-biological systems). It provides access to facilities forcomputational modelling where geographically separated scientists require accessto computational models, specialized parameter databases, management ofdistributed computing power and visualization of experimental data. Informationarchitectures included in the Virtual Kidney provide integration of parameterand evidence-based databases, computational models, text-mining tools anddistributed computing solutions with flexible and intuitive real-world interfacesand visualization options. These features will support developments in virtualexperimental design and are likely to be of significant value in the development ofclinical decision support environments.

The Virtual Kidney project is concerned with the development of aninteractive distributed web-based repository of models of the mammaliankidney. An interactive portal incorporates a three-dimensional anatomicalimage of a generic kidney with a multi-scale representation that provides atool for browsing and access to a collection of distributed published models at alllevels of renal physiology. This enables researchers to integrate and combineexisting and new models, and non-modellers to interact with the models, alteringkey parameters according to their own hypotheses and visualizing the simulationresults in a variety of formats. The portal will include the existing types ofmodels relevant to renal physiology, including kinetic models of transporters andchannels, transport models of individual cell types, flat model epithelia (such asbladder and cultured epithelia) and tubular segments along the nephron, modelsof the microcirculation, of tubuloglomerular feedback and of inner and outermedulla at various levels of detail. Importantly, the portal will provide access toalternative models for various mechanisms, which may be based on different

Phil. Trans. R. Soc. A (2009)


2143The Virtual Kidney: an eScience portal


interpretations of the experimental data. To date, it has been more usual for anexperimenter or modeller to select a single preferred model because it has beendifficult to interact with alternative versions that have been developed in otherlaboratories using different programming environments.

Further development of the functional and structural kidney models will takeinto account the variety of scales, numerical methods and programminglanguages used in kidney modelling studies, and implementation of the resourcewill be facilitated by translation of the models into established biological systemmarkup languages such as CellML and FieldML (Lloyd et al. 2004). For proof ofconcept, we have implemented a simulation of a particular kidney function,namely regulation of fluid and electrolyte transport, by modelling transportwithin a specific cell type (distal tubule cells of the early distal convoluted tubule(DCT)). The implemented model concerns only distal tubule cells of the earlyDCT; later models, to be added during planned enhancements, add principal andintercalated cells and integrate the cell model into a length of distal tubules.Further extensions will include implementation of cellular transport models in allsegments in a single nephron, and then as a parallel system representing thewhole kidney.

The major resources and connectivity associated with the current implemen-tation of the Virtual Kidney are shown in figure 1.

(b ) eScience portals

An eScience portal is simply a website or desktop tool for accessing andrunning scientific applications and data, or linking together such applicationsinto tool chains. Such portals have several different flavours, which are as follows.

—Desktop utilities. Graphical user interfaces (GUIs) for wrapping command-lineapplications, which allow users to launch and run applications on a remotesupercomputer using a ‘drag and drop’ style GUI that hides complexity andplatform dependencies. Many of these have been developed, e.g. as part of theUSA ‘Science Portals’ initiative, but adoption has been limited as they do notoffer significant productivity gains and often hide or limit access to applications.

—Workflow wrappers. Building a ‘wrapper’ around a chain or collection of tools toautomate common operations (e.g. KEPLER, or more recently in the bioinfor-matics domain, BIOCONDUCTOR or GENEPATTERN). These tools are becoming morecommon or popular in cutting-edge bioinformatics research, as they automate orsimplify the process of running chains of tools (e.g. a mass spectrometry tool forproteomics, followed by a statistical analysis tool followed by a link to a databasetool for genetic information).

— Integrative research platforms. Tools that allow users to combine and/orexplore several tools, data observations and research. Such tools are far lessmature and are open research areas. The Kidneyome portal described in thispaper is such a tool.

Research can be classified as experimental/observational, modelling/simulation or integrative, where integrative means combining multipleobservations/experiments and models developed by different research teams.It is clear that integrative research is becoming increasingly important as new



researcher

experimentplan

eResearch Grid portal

user manager

resource manager

data manager

virtual organization

Grid scheduler (GRIDBUS broker)

kidney modelling

visualization

geographically distributed computational resources

1 2

…

n

model model

Figure 1. Overview of the resources and connectivity involved with operation of the Virtual Kidney(Kidneyome). A three-dimensional interactive interface allows users to explore the anatomy of thekidney and provides access to the quantitative kidney database (QKDB). A Grid portal presents auser interface to distributed kidney models enabling researchers to integrate and combine existingand new models, and non-modellers to alter key parameters and visualize simulation results.



insights and knowledge build on the increasing wealth of observational data andlegacy models (Stein 2008). Integrative research is highly dependent on softwareplatforms and it is important to distinguish between software platforms andconventional software tools, such as spreadsheets or toolkits such as Matlab.Software platforms are designed to seamlessly integrate the existing toolstogether, and are specifically designed to meet the needs of a researchcommunity, unlike general purpose tools such as Matlab.

The high-level generic architecture of an eResearch platform to supportintegrative Physiome research is illustrated in figure 2.

The ‘front end’ or user interface to an eResearch platform for Physiomeresearch is typically a web portal: a website that provides information fromdiverse sources in a unified way. A generic eResearch portal can provide thefollowing range of services and middleware.

—A ‘user manager’ for authentication, and managing a user’s environment andpreferences (often integrated into the portal).

—A ‘virtual organization (VO) manager’ to allow users to form and join projectsto share resources such as observational data or models on a per-project basis(Alfieri et al. 2003).



extended

kidneymodel

registry

Melbourne, AustraliaMelbourne, Australia

MelbourneGrid server

GRIDBUS resourcebrokereResearch Grid portal

the 3Dinterface

Melbourne, Au

kidneymodel 1

evryl

France

experimentplan

developed

developed

developed

developed

resultsvisualization on desktop

VO andauthorization

service

quantitativekidney

database

advancedvisualization

facilityVPAC,

Melbourne

resultarchive

withmeta-data

distaltubulemode

Auckland, NZ

developed developed

developed

Figure 2. The high-level generic architecture of a Grid-based eResearch platform to supportintegrative Physiome research.



—A ‘resource manager’ managing the top-level resources available to a user,including resource discovery processes such as ‘metadata services’ andproviding wrappers to legacy services.

—A ‘data manager’ to provide access to Grid data sources through interfacestandards such as STORAGE RESOURCE MANAGER (http://sdm.lbl.gov/srm-wg/)or brokers such as STORAGE RESOURCE BROKER (Baru et al. 1998).

—A ‘Grid scheduler’ running remote computational models and assemblingthe results. An example relevant to this paper is GRIDBUS (Venugopalet al. 2004).

Underlying most current eResearch projects is the notion of a ‘Grid’ (Foster &Kesselman 1999), or underlying infrastructure and middleware that connecttogether compute and data infrastructure across a wide area network, forexample the Globus TOOLKIT and middleware (Foster & Kesselman 1998).

More recently, the concept of a Grid has been superseded by ‘cloudcomputing’, which is a Grid in which the physical connections are hidden fromthe user and a user can assemble a cloud of resources and tools. Cloud computing


http://sdm.lbl.gov/srm-wg/




simply means that the integrative research platform must hide Grid tool detailsand dependencies from users, and conversely that the Grid tools, such asPBS and Condor in figure 1, must support interoperation better.

(c ) Mathematical modelling for the kidney

Mathematical modelling has been used for many years in analysing theoperation and function of complex biological systems and, in particular,the approach has been very successful in modelling heart physiology rangingfrom the molecular level to the whole organ (Noble 2002; Hunter & Nielsen 2005).Computational modelling has long played a key role in research on kidneyphysiology, and research groups around the world have constructed a wide varietyof mathematical models in different formats and simulation environments. Thesemodels have been useful tools for their developers but have often been proprietarywith limited interoperability, restricting access to other users and providingbarriers to the integration of models derived from different sources.

The World Wide Web coupled with distributed computing resources providesopportunities to integrate and explore renal models in a distributed modelrepository, allowing researchers a ready and interactive access to a collection ofmodels and resources that minimize the requirement for specialist programmingor computational modelling expertise on the part of the experimenter. Suchknowledge systems or ‘cyberinfrastructures’ are becoming available in manyareas of eScience (Stein 2008), and it is timely to consider their development anddeployment in the areas of physiology and related life sciences.

The direction of mathematical modelling in research on the physiology of renaltransport systems has been determined by the complex interactions amongcoupled flows, kidney metabolism and detailed anatomy at all levels of kidneyorganization, and also by the inaccessibility of the inner kidney structures toin vivo intervention. The behaviour in vivo of nephron segments and bloodvessels ‘hidden’ within the kidney interior must be inferred from micropuncturesamples taken at accessible surface sites, from measurements performed in vitroby such techniques as microperfusion, vesicle studies, patch clamp, immuno-fluorescence imaging and from the anatomical/architectural features of thekidney. Theoretical analyses have been crucial to quantitative formulation ofworking hypotheses, interpretation of experimental results and development ofnew experimental techniques required for the measurement of crucial parameterswhose importance was appreciated during modelling studies. For example,modelling studies have helped our understanding of glomerular filtration,(quasi)isotonic reabsorption in the proximal tubule, the roles of several nephronsegments in acid–base regulation and the mechanisms involved in autoregulationand tubuloglomerular feedback. The importance of countercurrent flows inproximal tubule fluid reabsorption and in the many ‘cycles and separations’relationships among medullary nephron segments, collecting ducts andmedullary blood vessels, and details of medullary microcirculation in relationto solute and water recycling have also been the subjects of extensive modellingwhere measurements have been difficult or impossible.

The models targeted for implementation in the core collection cover a widerange of scales and represent some of the classic studies in the field. The initialimplementation of three example models is reported in this paper and it is





planned to revise the range of interaction options and the interface in response tofeedback from users. Authors of other models will be encouraged to assist withthe preparation of their models for inclusion in the Virtual Kidney. The currentlist of examples includes (but will not be limited to) the following:

(i) proximal tubule reabsorption (Weinstein 1986, 1994, 1998a,b; Thomas &Dagher 1994);

(ii) distal tubule reabsorption (Chang & Fujita 1999);(iii) tubuloglomerular feedback (Holstein-Rathlou & Marsh 1990; Holstein-

Rathlou et al. 1991; Layton et al. 1991);(iv) glomerular filtration (Deen et al. 1972);(v) medullary models of the urine-concentrating mechanism

—simple central core model (Stephenson et al. 1976, 1987; Foster &Jacquez 1978);

—multi-individual-nephron flat model (Lory 1987; Thomas 1991);— shunted-multi-nephron flat model (Hervy & Thomas 2003);— innermedullary collectingduct acid–base transport (Weinstein1998a,b); and

(vi) a library of channel and transporter kinetics models.

2. The Virtual Kidney: an eScience portal

A web-based eScience portal has been developed for access to experimental dataand parameter values abstracted within a quantitative kidney database (QKDB)and to a repository of mathematical models of kidney function.

There are three specific aims for the project.

(i) Workflow automation. To implement the infrastructure necessary forinteracting with the core set of remote modelling resources on diversecomputing platforms, and managing this Grid-based collaborative onlineresource in terms of communication protocols, user management, sessionmanagement, authentication, resource registration, etc.

(ii) Visualization and data fusion. To develop software for converting andrendering diverse result sets from the core set of legacy models in acommon web-based three-dimensional interface.

(iii) Model integration. To adapt the development-intensive interface tools tothe particular problems posed by kidney modelling so that we can proceedwith integration of many more models at all scales with only minimalchanges in the interface software. The ‘workflow’ is not rigid or fixed andnew models can be readily added and/or adapted. This is the emergingtrend in bioinformatics and is a feature of tools, such as GENEPATTERN,which allows a flexible integration of genetic and proteomic analysis.

(a ) Three-dimensional portal

The portal presents a cross-disciplinary approach to providing researcherswith access to computational models developed in laboratories around the worldand uses global distributed computing to manage workflow and deliver solutionsto the user’s desktop.





The user is presented with a three-dimensional ‘Virtual Kidney’ interface,based on a translucent multi-scale reconstruction of a rat kidney showing majoranatomical structures as well as renal arteries and veins to several levels ofbranching derived from synchrotron images (Nordsletten et al. 2006). Within thekidney are examples of reconstructed serially sectioned nephrons. The user canrotate the kidney, display or hide individual structures and zoom in on anyportion. The selection of a structure provides a link to the search facility of aspecialized database, the QKDB, which contains a comprehensive catalogue ofparameter values, extracts from the literature and other relevant resources, suchas collections of anatomical and histological images.

The basic principle is to create a web page that presents the user (i.e. thevisitor to the website; also called the ‘client’) with an annotated list or a visualinterface for selection of one of the available models. Once this choice has beenmade, the user is presented with a more detailed description of the chosen model,including literature references, and is given the opportunity either to look at theresults of some pre-configured simulations to ‘get the feel’ of the model, or toalter particular parameter values (within a constrained range representingphysiologically reasonable values) in order to customize the model as required.Upon submitting the new parameter values, the user will be served, after a waitthat depends on the execution time of the model calculations, with the resultsand may choose among several styles of presentation. These include simple tablesor graphical output, or in some cases more visually appropriate presentations, orthe user may request to download the results as a table for post-treatment in aspreadsheet or graphing program of their own. An important feature of the portalis the ability to save ‘session state’, such that users can return to an interactivesession and take up where they left off.

The screenshots shown in figure 3 are derived from an early version of theVirtual Kidney interface (Lonie et al. 2004). This preliminary model wasconstructed from micro-CT scans of a rat kidney and serially sectioned casts ofmouse kidney tubules. The model is designed not only for exploring kidneyanatomy but also for allowing a selection of structures with linked access toQKDB and for the collection of curated, interactive models at various scales.When running some simulation models, spatial distributions of quantitativeoutput data derived from calculations from models can be mapped onto theanatomical representation in various ways. For instance, solute concentrationsderived from a nephron flow model can be mapped back to the nephrongraphically using look-up tables to colour code ranges of concentration andassigning the relevant colour to voxels representing the location of solutions ofthis concentration (figure 3, right side). This GUI is built on the XML UserInterface Language environment of the Mozilla open source browser. It iseffectively a web-based version of a stable and proven three-dimensionalvisualization environment developed by the Bioengineering Institute in Auckland,which has been used extensively for cardiac physiology simulation and cardiacanatomy navigation (Christie et al. 2002). The Melbourne-based members ofour group have been working closely over the past few years with the BioengineeringInstitute to develop the web-based version of this environment, and enhancethe environment with the specific functionality required for rendering renalmodelling results. This work has also addressed the need to develop the softwaretools required for converting and rendering diverse sets of kidney modelling results



Figure 3. Screenshot of the Virtual Kidney interface showing (left side of screen) the three-dimensional anatomical representation of a kidney with vasculature and example nephrons,derived from reconstructed serial sections and (right side of screen) graphical output of a distaltubule simulation (Chang & Fujita 1999) in which selected solutes (shown in colour) are plotted asconcentrations against distance along tubule segment.



in this common three-dimensional interface—for instance, converting andrendering the results depicted in the kidney simulation (KSIM) interface depicted infigure 3. The Virtual Kidney interface is intended both as a means of navigatingthe model databases and as a means of running model simulations and viewingsimulation results, with the only requirement being a Mozilla-based browser.

(b ) Quantitative kidney database

Our QKDB (Dzodic et al. 2004; Ribba et al. 2006) is open via a web GUI(http://physiome.ibisc.fr/qkdb/) and is in use among a small group ofcollaborating laboratories. It is built on a flexible, extensible and generic datamodel (entity-relationship model) implemented under MYSQL/PHP/APACHE.This database has been seeded from the bibliography resources of the presentproject participants and presently includes experimental data from some300 research publications. QKDB is open not only for consultation but, equallyimportantly, for (password-protected) contributions from the renal researchcommunity. This resource thus puts legacy measurements, as well as recent andnew data, at the ready disposal of renal researchers, thereby facilitating


http://physiome.ibisc.fr/qkdb/




comparisons of results obtained in different species and under variousexperimental conditions. It includes the following types of information, forhuman kidneys where known, but especially in experimentally studied species(mammalian, amphibian and avian) and in model epithelia such as cultured cellsand amphibian skin and urinary bladder:

— transport parameters, such as permeability to water and various solutes,kinetics of transporters and channels, in all nephron and blood vessel segmentsand kidney regions;

— tubular concentrations and flow rates along the various segments of thenephron and blood vessels;

—qualitative and quantitative anatomical details, such as tubule diameters andepithelial and cellular dimensions for the various nephron and blood vesselsegments; relative placement of structures in each kidney region; typicalkidney sizes and weights for different species; dimensions; and

—cortical and medullary regions and subregions, and anatomical images, etc.

Importantly, the experts who enter the data via the web interface are encouragedto include annotations concerning experimental conditions, relevance andlimitations of experimental techniques, etc. Quality control of the curated data isassured at several levels: in addition to syntax checking on entry form contents andthe use of pull-down menus built dynamically from the contents of the database(thus avoiding duplication and multiple spellings of field names), the data areentered only by authenticated researchers, under password control. Finally, a boardof experts oversees new entries before they become accessible to web queries.

(c ) Model repository and links

The portal provides an interactive user interface to a collection of distributedpublished models at all levels of renal physiology, enabling researchers tointegrate and combine existing and new models, and non-modellers to interactwith the models, altering key parameters according to hypotheses of their ownand visualizing the simulation results. The portal will include all existing types ofmodels relevant to renal physiology, including kinetic models of transporters andchannels, transport models of individual cell types, of flat model epithelia (suchas bladder and cultured epithelia) and of tubular segments along the nephron,models of the microcirculation, of tubuloglomerular feedback and of inner andouter medulla at various levels of detail. Importantly, the portal will provideaccess to alternative models for various mechanisms, which may be based ondifferent interpretations of the experimental data. To date, it has been moreusual for an experimenter or modeller to select a single preferred model since ithas been difficult to interact with alternative versions that have been developedin other laboratories using different programming environments.

The underlying models are implemented via XML markup descriptions(e.g. CellML, see www.cellml.org) or as executable program files stored locally onthe server or remotely (e.g. on the model author’s server). We have soughtapproaches that are consistent, as far as possible, with standards and ontologiesbeing developed for other organs under the Physiome initiative.


http://www.cellml.org




A menu lists the repository of mathematical models, such as segmental(e.g. proximal or distal tubule), regional (e.g. medullary transport) or multi-nephron complex systems models, which may be run on the local machine or, viaa Grid portal (KIDNEYGRID), simultaneously on several remote machines. Thesimulation results (e.g. solute concentrations or local flow rates) are visible onthe Virtual Kidney structures as scaled colour gradients, thus allowing a visualand quantitative appreciation of the effects of simulated parameter changes.Figure 4b shows x–y plots of the results. The selection of a progressive set ofparameter changes allows systematic evaluation of the effects of changes in oneor more elements of a model and can be used to predict the probable outcomes ofexperimental interventions. This approach is likely to become an increasinglyimportant aspect of experimental design testing and could have significantbenefits in determining the cost-effectiveness of experimental programs.

The portal allows scientists to explore from their desktop a variety ofcomputational models that have previously been accessible only within thelaboratory in which they were developed. It is important to note that the listedlegacy kidney models and resources represent a number of different implemen-tation techniques (different modes of interaction—command line, graphicalinterface and command file driven—running on different platforms). Aninteroperability problem arises during attempts to integrate or co-locate suchlegacy models because standards have not been available or adopted for datarepresentation, interaction or communication.

There are several ways of addressing this issue, the best of which might be torecode all the models in a common representation format as discussed above.However, this is obviously a long-term and resource-intensive solution, andprobably not feasible without a direct and substantial input from each of themodel developers.

We have adopted a reasonable alternative approach, namely to developinteraction ‘wrappers’ around the current model implementations and convertthe output in the central portal so that diverse result sets can be rendered on acommon interface. This approach requires infrastructure ‘middleware’ to managethe communication channels, data transfer channels, user session state, userauthentication, etc., within the context of a central portal.

(d ) Model examples

(i) Distal tubule reabsorption and secretion

The implementation is based on the model of Chang & Fujita (1999);currently, we have two cell types working with advection–diffusion spatialmodelling and have demonstrated agreement of the numerical output with thatin the published work. This has been implemented and currently provides theluminal output results. The intracellular and other output results documentedin the original publication have not been included as yet. In the longer term,the aim of the Kidneyome project is to develop a platform for ‘plug and play’modelling, where there is a standard basis of representation for all models(preferably an XML-based open modelling markup language such as CellML; seewww.cellml.org) and data representation is similarly standardized. We have madeinitial proof-of-concept moves towards this goal by developing a CellML-based


http://www.cellml.org




implementation of the Chang and Fujita model (although the input and outputdata formats are still ‘proprietary’ to some extent) and rendering the results in anopen standards-based three-dimensional interface. The next stage will involveintegrating the transport characteristics of the various distal tubule cell types tomodel fluid dynamics and ionic fluxes along this tubule segment in the context ofan accurate structural nephron model. Detailed models of distal tubule transportare of interest because gain-of-function mutations in distal tubule transporters(e.g. the thiazide-sensitive NaCl cotransporter) or ion channels (e.g. epithelialNa channel) involved in reabsorption of sodium chloride are implicated in severalforms of hereditary hypertension (Lifton et al. 2001; Tripodi et al. 2004).

Future extensions will include a library of kinetic models of the differentchannels and transporters (and their various isoforms) expressed in all cell typesalong the nephron and implementation of cellular transport models in allsegments in a single nephron, and then as a parallel system representing thewhole kidney.

(ii) Medullary models of the urine-concentrating mechanism

Shunted multi-nephron flat model (Hervy & Thomas 2003) KSIM PRESENT

STATE. A prototype Java applet (http://www.ibisc.univ-evry.fr/wsrthomas/kidneysim/) has been used to display simulation results, and its use isdemonstrated for one of the models of the renal medulla (Hervy & Thomas2003). On this website, separate web pages present: the history leading up to thedevelopment of this particular model; a description of the model itself, includinga table of the basic parameter values; and finally a window giving entry into aJava applet for presentation of simulation results. The visitor may choose amonga small set of previous simulations (typical of those presented in the publishedpaper) or may launch a new simulation (server-side execution) based on theirmodifications of a selected set of model parameter values. The applet loads thechosen set of simulation results in the form of an XML file and permits the webvisitor to display them as x–y plots or as a colour gradient diagram of themedullary structures (figure 4).

Development of this demonstration provides valuable experience concerningthe possibilities and limitations of existing graphics packages for Java appletsand reveals key development features for development of a more flexibleand generic approach for use with the variety of models projected for theKidneyome site.

(iii) Multi-nephron model

In addition to segmental and medullary transport models, the portal currentlysupports a multi-nephron model as demonstrated by the inclusion of a dynamicalnetwork model (Moss et al. 2009). The modelling approach combines networkautomata, graphs in which each node has a state and a set of rules for changingthe state, into a single model. The structure of the model is explicitly representedas a network of graph automata, which captures the hierarchy, interactionsand couplings in the system. The model captures the dynamics of a system ofnephrons by maintaining a system state at each node of the network, and througha set of update rules that operate on the state at each node. This approach is wellsuited to studying emergent dynamics—dynamics that arise from the


http://www.ibisc.univ-evry.fr/~srthomas/kidneysim/

http://www.ibisc.univ-evry.fr/~srthomas/kidneysim/




interactions and couplings between nephrons—at different levels of hierarchy.The purpose of this model is to explore the ways in which the stability of whole-kidney function arises from the dynamics of individual nephrons.

The nephron tubule is modelled as one-dimensional graph automata,consisting of one node for each functionally distinct nephron segment and onenode for each layer of interstitial medullary fluid in which the nephron issituated. Larger systems of nephrons are built up by connecting with edgesmultiple nephron models to the same interstitial fluid nodes. Whole-kidneymodels can be constructed by mimicking the renal pyramid structure of thekidney and collecting duct systems and relating regions of interstitial fluid to anarterial tree. The arterial tree is modelled as a hierarchical network, whichdistributes blood to each nephron automatum in proportion to the resistances ofthe pre-glomerular blood vessels including the afferent arterioles.

Interactions between nephron segments and between nephrons andthe interstitial environment are achieved via the edges of the network and theupdate rules that operate on the nodes of the graph. Nephron dynamics arecaptured by discrete time and difference equations. Solutions are obtained fornew inputs or disturbances by combining multiple equations to reach a sharedequilibrium; for example, as in the diffusion of water within multiple descendinglimbs of Henle. Equations are solved simultaneously using Gaussian elimination.

The model has been used to study emergent dynamics, and the interactionsthat arise between haemodynamic coupling and vascular signalling, in thecontext of two-, eight- and 72-nephron systems. The experiments performed todate have focused on the dynamics of systems of nephrons, rather than individualnephrons or nephron segments. Specifically, the experiments have exploredchanges in the filtration rates of entire columns of nephrons, and the filtrationrate of the system as a whole and exploring the power spectral densities todetermine the dominant frequencies present in these filtration rate values. Theexperiments demonstrated that the degree of vascular signalling present betweennephrons in the system affects both the magnitude of their filtration rates andalso which frequencies are dominant.

Further experiments using this model have studied the emergent effects oflocalized perturbations of selected input parameters on whole-system behaviour.In a typical analysis, the reabsorption of water in the descending limb of Henlewas impaired for a fraction of the nephrons in the system, and the effects on thewhole-system filtration rate were observed. These experiments consistentlydemonstrated that the system dynamics were highly stable and were only slightlyaffected by the impairments. These experiments were the first study of theemergent dynamics produced by impaired nephron function, and serve toillustrate how the model may be used to predict emergent dynamics induced bylocalized renal disease. An example of the use of the portal to select inputparameter data and to display the output from the model is shown in figure 5.

This modelling approach has been demonstrated to scale up to much largersystems than existing multi-nephron models (Marsh et al. 2007). Based on timingmeasurements and a performance analysis of the model implementation, thesimulation of whole-kidney function based on the dynamics of individualnephrons is feasible for the first time. A simulation can also be distributedover multiple computers, resulting in a performance gain of up to a 10-foldspeed up when compared with a single computer simulation. In order to simulate



(a)

(b)

Figure 4. Medullary model of the urine-concentrating mechanism. (a) Example result showingdistribution of urea concentration within the medulla, displayed using a colour gradient torepresent concentration values. (b) Graphical output from the model representing changes inosmolarity along the length of the long descending limb of the loop of Henle (ldl).



whole-kidney function, a model of an entire kidney (e.g. the human kidneycontains approx. 1 million nephrons) is required and since the network model iscompositional, it is possible to grow a whole-kidney model.

Using the network model, experimenters can also explore the emergentdynamics of multi-nephron or whole-kidney models. Various parameters can becontrolled by the experimenter, such as the degree of haemodynamic couplingand vascular signalling in the system, the filtration rates of the individualnephrons and various parameters governing fluid and solute transport. More



(a)

(b)

Figure 5. Example of the use of the portal to run the multi-nephron model. (a) Input parametersscreen allows the choice of two-nephron or eight-nephron model and provides for selection ofparameter scan within a range or over a series of designated time steps with single values for eachparameter under investigation. (b) Example results from two-nephron model (red or greenrepresenting each nephron in the pair). When plotted against time, single nephron glomerularfiltration rate in each nephron is shown to oscillate out of phase with the other.



importantly, the network allows us to replace any of the currently used updaterules with equations from other models. The systems that are available forexperimentation through the portal are currently limited to two-nephron andeight-nephron models but models of up to 512 nephrons have been executed.

(e ) Grid portal

The use of a Grid resource broker demonstrates the ability to compose,schedule, monitor and visualize the results of simulations and simplifies thedevelopment of application, credential and resource management, while





decoupling the launch platform from the underlying Grid middleware. Thedevelopment, architecture, features and performance of the KIDNEYGRID platformincluded within the Virtual Kidney have been described by Chu et al. (2008).The steps involved in running a Grid-enabled session are the following.

(i) User initiates session with portal by selecting modelling resourcesrequired and ‘planning’ experiment with the resources available.

(ii) User completes authorization through Globus middleware infrastructure.(iii) Portal contacts Grid broker to request resources (Remote/Local Kidney

Database or remote simulations).(iv) Grid broker queries registry to provide reference to resource. Registry

returns references to resources required and broker returns reference plusany important control data to the portal.

(v) Broker opens communication channel to remote resource and sets up.Note that resources may be locally based (e.g. in the local mirror of theKidney database) or hosted on remote servers.

(vi) More commonly, simulations will be hosted on remote sites, running onproprietary systems. The Grid broker will manage communications withthe remote resource through a combination of tailored communicationsand/or implementation of service-based wrappers at the remote resource.It is preferable that as few resources as possible are required to beimplemented on the remote resource, and communication brokering willtherefore most probably be tailored to individual model resources withmost processing of result datasets occurring at the eResearch portal.

(vii) User session state, including the current state of experiment, is saved tolocal database.

(viii) Results are visualized on the user’s desktop in their browser that isrunning the three-dimensional Virtual Kidney portal.

(ix) Alternatively, for complex datasets, it may be preferable to use advancedvisualization resources, such as the Victorian Partnership for AdvancedComputing Advanced Visualization Facility in Melbourne.

3. Future developments

A long-term challenge for the Physiome initiative is to build a modellingframework in which the effect of a gene mutation can be modelled all the wayfrom its effect on protein structure and function to how the altered properties ofthe protein affect a cellular process such as signal transduction, and how thechanged properties of that process alter the function of tissues and organs. It willalso help to clarify the way in which environmental influences, along with geneticexpression, influence physiological function. There will be many other benefitsfrom this integrative framework. Understanding how model parameters areaffected by individual variation, embryological growth, ageing and disease, forexample, will have enormous potential benefits for the design of medical devices,the diagnosis and treatment of disease and the development of new drugs.Extension of such models to explore the effects of pathology and genetic disordersis now possible with the growth of computing power and Internet connectivityand ready access to such resources through eScience portals will be a centralpillar of the VPH endeavour.





The Kidneyome or Virtual Kidney will contribute to the overall generation ofa set of computational resources relevant to all aspects of human physiology andpathology. In the short term, it is expected that the portal will accumulaterelevant curated data from the literature and provide experimental scientistswith access to a wide variety of models with which they can simulate effects ofchanges in their chosen renal system, and thus inform the process ofexperimental design. At this stage, we have included only a few long-standing‘legacy’ models, some of which are now outdated. We plan to add more currentmodels including a reimplementation of some distal tubule models (Weinstein1994, 1998a,b). The concentrating mechanism is more problematic as there aresimply not any ‘successful’ models since there is no satisfactory explanation of allaspects of the mechanism. More value will be added to the portal by including alibrary of channel and transporter kinetic models as they become available and,we anticipate, models of the various cell types along the nephron.

In the longer term, the integration of various low-level models with systems ornetwork approaches is likely to allow the development of interactive multi-scalerepresentations of the kidney that can be placed within the other body systems toexplore quantitatively the activity of control systems involved in the regulationof the cardiovascular system and fluid and electrolyte balance.

The Physiome Project relies on software tools to facilitate constructing andexploring models, but such tools are unlikely to ever be a single unified model;instead, our view is that the Physiome will be enabled by a range of eScienceportals, such as the Virtual Kidney, customized for the researchers and researchissues for specific organs and systems, ranging from kidneys to the respiratory orlymphatic systems.

S.R.T. gratefully acknowledges financial support from the French National Research Agency (ANRgrants ANR-06-BYOS-0007-01 (SAPHIR) and ANR-05-BLAN-0247-06 (CorBioMath)), theEuropean Community’s Seventh Framework Programme (FP7/2007-2013, grant agreement no.223920, VPH-NoE), the Council of Essonne Region (Pole System@tic, POPS project) and GdRSTIC-Sante (CNRS 2647 and INSERM). P.J.H., S.R.T., P.J.H., R.B. and W.A. received fundingfrom the Australian Research Council (Special Research Initiatives: e-Research grant), and theVictoria Partnership for Advanced Computing.

References

Alfieri, R., Cecchini, R., Ciaschini, V., dell’Agnello, L., Frohner, A. & Gianoli, A. 2003 VOMS: anauthorization system for virtual organizations. In First European Across Grids Conference,Santiago de Compostela.

Baru, C., Moore, R., Rajasekar, A. & Wan, M. 1998 The SDSC storage resource broker. In Proc.CASCON’98, Toronto, Canada.

Bassingthwaighte, J. B. 2000 Strategies for the Physiome Project. Ann. Biomed. Eng. 28,1043–1058. (doi:10.1114/1.1313771)

Chang, H. & Fujita, T. 1999 A numerical model of the renal distal tubule. Am. J. Physiol. 276,F931–F951.

Christie, R., Bullivant, D., Blackett, S. & Hunter, P. J. 2002 Modelling and visualising the heart.Comput. Vis. Sci. 4, 227–235. (doi:10.1007/s00791-002-0079-3)

Chu, X.-C., Lonie, A., Harris, P. J., Thomas, S. R. & Buyya, R. 2008 A service-oriented Gridenvironment for integration of distributed kidney models and resources. Concurrency Comput.Pract. Exp. 20, 1095–1111. (doi:10.1002/cpe.1285)


http://dx.doi.org/doi:10.1114/1.1313771

http://dx.doi.org/doi:10.1007/s00791-002-0079-3

http://dx.doi.org/doi:10.1002/cpe.1285




Deen, W. M., Robertson, C. R. & Brenner, B. M. 1972 A model of glomerular ultrafiltration in the

rat. Am. J. Physiol. 223, 1178–1183.Dzodic, V., Hervy, S., Fritsch, D., Khalfallah, H., Thereau, M. & Thomas, S. R. 2004 Web-based

tools for quantitative renal physiology. Cell. Mol. Biol. 50, 795–800.Fenner, J. W. et al. 2008 The EuroPhysiome, STEP and a roadmap for the virtual physiological

human. Phil. Trans. R. Soc. A 366, 2979–2999. (doi:10.1098/rsta.2008.0089)Foster, D. M. & Jacquez, J. A. 1978 Comparison using central core model of renal medulla of the

rabbit and rat. Am. J. Physiol. 234, F402–F414.Foster, I. & Kesselman, C. 1998 The Globus Project: a status report. In Proc. IPPS/SPDP’98

Heterogeneous Computing Workshop, pp. 4–18.Foster, I. & Kesselman, C. 1999 The Grid: blueprint for a future computing infrastructure. San

Francisco, CA: Morgan Kaufmann Publishers.Hervy, S. & Thomas, S. R. 2003 Inner medullary lactate production and urine-concentrating

mechanism: a flat medullary model. Am. J. Physiol. Renal Physiol. 284, F65–F81.Holstein-Rathlou, N. H. & Marsh, D. J. 1990 A dynamic model of the tubuloglomerular feedback

mechanism. Am. J. Physiol. 258, F1448–F1459.Holstein-Rathlou, N. H., Wagner, A. J. & Marsh, D. J. 1991 Tubuloglomerular feedback dynamics

and renal blood flow autoregulation in rats. Am. J. Physiol. 260, F53–F68.Hunter, P. J. & Borg, T. K. 2003 Integration from proteins to organs: the Physiome Project. Nat.

Rev. Mol. Cell Biol. 4, 237–243. (doi:10.1038/nrm1054)Hunter, P. J. & Nielsen, P. M. F. 2005 A strategy for integrative computational physiology.

Physiology 20, 316–325. (doi:10.1152/physiol.00022.2005)Layton, H. E., Pitman, E. B. & Moore, L. C. 1991 Bifurcation analysis of TGF-mediated

oscillations in SNGFR. Am. J. Physiol. 261, F904–F919.Lifton, R. P., Gharavi, A. G. & Geller, D. S. 2001 Molecular mechanisms of human hypertension.

Cell 104, 545–556. (doi:10.1016/S0092-8674(01)00241-0)Lloyd, C. M., Halstead, M. D. B. & Nielsen, P. M. F. 2004 CellML: its future, present and past.

Prog. Biophys. Mol. Biol. 85, 433–450. (doi:10.1016/j.pbiomolbio.2004.01.004)Lonie, A. J., Stevens, C. & Harris, P. J. 2004 Computer modeling of kidney function. In Proc. Exp.

Biol. LB520, Washington, DC.Lory, P. 1987 Effectiveness of a salt transport cascade in the renal medulla: computer simulations.

Am. J. Physiol. 252, F1095–F1102.Marsh, D. J., Sosnovtseva, O. V., Mosekilde, E. & Holstein-Rathlou, N.-H. 2007 Vascular coupling

induces synchronization, quasiperiodicity, and chaos in a nephron tree. Chaos 17, 015 114.

(doi:10.1063/1.2404774)Moss, R., Kazmierczak, E., Kirley, M. & Harris, P. 2009 A computational model for emergent

dynamics in the kidney. Phil. Trans. R. Soc. A 367, 2125–2140. (doi:10.1098/rsta.2008.0313)Noble, D. 2002 The rise of computational biology. Nat. Rev. Mol. Cell Biol. 3, 459–463. (doi:10.

1038/nrm810)Nordsletten, D., Blacket, S., Bentley, M., Ritman, E. & Smith, N. P. 2006 Structural morphology

of renal vasculature. Am. J. Physiol. 291, H296–H309. (doi:10.1152/ajpheart.00814.2005)Ribba, B., Tracqui, P., Boix, J. L., Boissel, J. P. & Thomas, S. R. 2006 QxDB: a generic database

to support mathematical modelling in biology. Phil. Trans. R. Soc. A 364, 1517–1532. (doi:10.1098/rsta.2006.1784)

Stein, L. D. 2008 Towards a cyberinfrastructure for the biological sciences: progress, visions andchallenges. Nat. Rev. (Genet.) 9, 678–688. (doi:10.1038/nrg2414)

Stephenson, J. L., Mejia, R. & Tewarson, R. P. 1976 Model of solute and water movement in thekidney. Proc. Natl Acad. Sci. USA 73, 252–256. (doi:10.1073/pnas.73.1.252)

Stephenson, J., Zhang, Y., Eftekhari, A. & Tewarson, R. 1987 Electrolyte transport in a central-

core model of the renal medulla. Am. J. Physiol. 253, F982–F997.Thomas, S. R. 1991 Effect of varying salt and urea permeabilities along descending limbs of Henle

in a model of the renal medullary urine concentrating mechanism. Bull. Math. Biol. 53, 825–843.(doi:10.1016/S0092-8240(05)80409-4)


http://dx.doi.org/doi:10.1098/rsta.2008.0089

http://dx.doi.org/doi:10.1038/nrm1054

http://dx.doi.org/doi:10.1152/physiol.00022.2005

http://dx.doi.org/doi:10.1016/S0092-8674(01)00241-0

http://dx.doi.org/doi:10.1016/j.pbiomolbio.2004.01.004

http://dx.doi.org/doi:10.1063/1.2404774




http://dx.doi.org/doi:10.1152/ajpheart.00814.2005



http://dx.doi.org/doi:10.1038/nrg2414

http://dx.doi.org/doi:10.1073/pnas.73.1.252

http://dx.doi.org/doi:10.1016/S0092-8240(05)80409-4




Thomas, S. R. In press. Kidney modeling and systems physiology. Wiley Interdisciplinary Reviews:Systems Biology and Medicine.

Thomas, S. R. & Dagher, G. 1994 A kinetic model of rat proximal tubule transport—load-dependent bicarbonate reabsorption along the tubule. Bull. Math. Biol. 56, 431–458.

Thomas, S. R. et al. 2008 SAPHIR: a Physiome core model of body fluid homeostasis and bloodpressure regulation. Phil. Trans. R. Soc. A 366, 3175–3197. (doi:10.1098/rsta.2008.0079)

Tripodi, G., Florio, M., Ferrandi, M., Modica, R., Zimdahl, H., Hubner, N., Ferrari, P. & Bianchi,G. 2004 Effect of Add1 gene transfer on blood pressure in reciprocal congenic strains of Milanrats. Biochem. Biophys. Res. Commun. 324, 562–568. (doi:10.1016/j.bbrc.2004.09.079)

Venugopal, S., Buyya, R. & Winton, L. 2004 A Grid service broker for scheduling distributed data-oriented applications on global grids. In Proc. 2nd Int. Workshop on Middleware for GridComputing. New York, NY: ACM Press.

Weinstein, A. M. 1986 A mathematical model of the rat proximal tubule. Am. J. Physiol. 250,F860–F873.

Weinstein, A. M. 1994 Mathematical models of tubular transport. Annu. Rev. Physiol. 56,691–709. (doi:10.1146/annurev.ph.56.030194.003355)

Weinstein, A. M. 1998a Insights from mathematical modeling of renal tubular function. Exp.Nephrol. 6, 462–468. (doi:10.1159/000020556)

Weinstein, A. M. 1998b A mathematical model of the inner medullary collecting duct of the rat—acid/base transport. Am. J. Physiol. 43, F856–F867.



http://dx.doi.org/doi:10.1016/j.bbrc.2004.09.079

http://dx.doi.org/doi:10.1146/annurev.ph.56.030194.003355

http://dx.doi.org/doi:10.1159/000020556


the virtual kidney: an escience interface and grid portal · the virtual kidney uses a web...

Documents